In [1]:
import time
import numpy as np
import pandas as pd
import itertools as it
from sklearn.model_selection import train_test_split, GridSearchCV, cross_val_score, learning_curve
from sklearn.neural_network import MLPClassifier
from sklearn.metrics import roc_curve, precision_score, accuracy_score, roc_auc_score
from sklearn.preprocessing import StandardScaler
import matplotlib.pylab as plt
from sklearn.manifold import TSNE
import plotly.graph_objects as go
standarscaler = StandardScaler()
from warnings import filterwarnings
filterwarnings('ignore')
In [2]:
Data = pd.read_csv('https://raw.githubusercontent.com/mcuscagua/Aprendizaje_Automatico/master/Trabajo_2/Data.csv')
Data = Data.set_index('Date')
Y = Data['Class']
X = Data.drop('Class', axis = 1)
T = TSNE(n_components=2).fit_transform(X)
X = pd.DataFrame(T)
X_train, X_test, Y_train, Y_test = train_test_split(X, Y, test_size=0.2)
X_train_Stand = standarscaler.fit_transform(X_train)
X_test_Stand = standarscaler.transform(X_test)
In [3]:
Neurons = [x+1 for x in range(10)]
all_combinations = [[] for k in range(len(Neurons))]
for i in range(10):
for j in range(10):
all_combinations[i].append(tuple((i+1 for k in range(j+1))))
In [4]:
TrSCr_LC_V02 = np.ndarray((len(all_combinations), 5, len(all_combinations[0])))
TeSCr_LC_V02 = np.ndarray((len(all_combinations), 5, len(all_combinations[0])))
AUC_v02 = np.zeros((len(all_combinations), len(all_combinations[0])))
Accuracy_v02 = np.zeros((len(all_combinations), len(all_combinations[0])))
TrSCr_LC_V05 = np.ndarray((len(all_combinations), 5, len(all_combinations[0])))
TeSCr_LC_V05 = np.ndarray((len(all_combinations), 5, len(all_combinations[0])))
AUC_v05 = np.zeros((len(all_combinations), len(all_combinations[0])))
Accuracy_v05 = np.zeros((len(all_combinations), len(all_combinations[0])))
TrSCr_LC_V09 = np.ndarray((len(all_combinations), 5, len(all_combinations[0])))
TeSCr_LC_V09 = np.ndarray((len(all_combinations), 5, len(all_combinations[0])))
AUC_v09 = np.zeros((len(all_combinations), len(all_combinations[0])))
Accuracy_v09 = np.zeros((len(all_combinations), len(all_combinations[0])))
In [5]:
start_time = time.time()
for i in range(len(all_combinations)):
for j in range(len(all_combinations[i])):
#--------- LR = 0.2 ---------#
MLPC = MLPClassifier(hidden_layer_sizes=all_combinations[i][j],
learning_rate = 'constant',
learning_rate_init = 0.2,
random_state=0)
MLPP = MLPClassifier(hidden_layer_sizes=all_combinations[i][j],
learning_rate = 'constant',
learning_rate_init = 0.2,
random_state=0)
MLPC.fit(X_train_Stand, Y_train)
y_score = MLPC.predict_proba(X_test_Stand)
y_pred = MLPC.predict(X_test_Stand)
train_sizes, train_scores, test_scores = learning_curve(MLPP, X, Y)
TSZ_LC_V02 = train_sizes
TrSCr_LC_V02[i,:,j] = np.median(train_scores, axis=1)
TeSCr_LC_V02[i,:,j] = np.median(test_scores, axis=1)
AUC_v02[i,j] = roc_auc_score(Y_test, y_pred)
Accuracy_v02[i,j] = accuracy_score(Y_test, y_pred)
#--------- LR = 0.5 ---------#
MLPC = MLPClassifier(hidden_layer_sizes=all_combinations[i][j],
learning_rate = 'constant',
learning_rate_init = 0.5,
random_state=0)
MLPP = MLPClassifier(hidden_layer_sizes=all_combinations[i][j],
learning_rate = 'constant',
learning_rate_init = 0.5,
random_state=0)
MLPC.fit(X_train_Stand, Y_train)
y_score = MLPC.predict_proba(X_test_Stand)
y_pred = MLPC.predict(X_test_Stand)
train_sizes, train_scores, test_scores = learning_curve(MLPP, X, Y)
TrSCr_LC_V05[i,:,j] = np.median(train_scores, axis=1)
TeSCr_LC_V05[i,:,j] = np.median(test_scores, axis=1)
AUC_v05[i,j] = roc_auc_score(Y_test, y_pred)
Accuracy_v05[i,j] = accuracy_score(Y_test, y_pred)
#--------- LR = 0.9 ---------#
MLPC = MLPClassifier(hidden_layer_sizes=all_combinations[i][j],
learning_rate = 'constant',
learning_rate_init = 0.9,
random_state=0)
MLPP = MLPClassifier(hidden_layer_sizes=all_combinations[i][j],
learning_rate = 'constant',
learning_rate_init = 0.9,
random_state=0)
MLPC.fit(X_train_Stand, Y_train)
y_score = MLPC.predict_proba(X_test_Stand)
y_pred = MLPC.predict(X_test_Stand)
train_sizes, train_scores, test_scores = learning_curve(MLPP, X, Y)
TrSCr_LC_V09[i,:,j] = np.median(train_scores, axis=1)
TeSCr_LC_V09[i,:,j] = np.median(test_scores, axis=1)
AUC_v09[i,j] = roc_auc_score(Y_test, y_pred)
Accuracy_v09[i,j] = accuracy_score(Y_test, y_pred)
elapsed_time = time.time() - start_time
print(elapsed_time)
Resultados¶
In [6]:
fig1 = go.Figure(data=[go.Surface(z=AUC_v02)])
fig1.update_layout(title='AUC Surface', autosize=True,
scene = dict(xaxis_title='Neurons',
yaxis_title='Layers',
zaxis_title='AUC'))
fig1.update_yaxes(tickvals=[x+30 for x in range(10)])
fig1.update_xaxes(tickvals=[x+30 for x in range(10)])
fig1.show()
In [7]:
fig2 = go.Figure(data=[go.Surface(z=AUC_v05)])
fig2.update_layout(title='AUC Surface', autosize=True,
scene = dict(xaxis_title='Neurons',
yaxis_title='Layers',
zaxis_title='AUC'))
fig2.update_yaxes(tickvals=[x+30 for x in range(10)])
fig2.update_xaxes(tickvals=[x+30 for x in range(10)])
fig2.show()
In [8]:
fig3 = go.Figure(data=[go.Surface(z=AUC_v09)])
fig3.update_layout(title='AUC Surface', autosize=True,
scene = dict(xaxis_title='Neurons',
yaxis_title='Layers',
zaxis_title='AUC'))
fig3.update_yaxes(tickvals=[x+30 for x in range(10)])
fig3.update_xaxes(tickvals=[x+30 for x in range(10)])
fig3.show()
In [9]:
fig4 = go.Figure(data=[go.Surface(z=Accuracy_v02)])
fig4.update_layout(title='Accuracy Surface', autosize=True,
scene = dict(xaxis_title='Neurons',
yaxis_title='Layers',
zaxis_title='Accuracy'))
fig4.update_yaxes(tickvals=[x+30 for x in range(10)])
fig4.update_xaxes(tickvals=[x+30 for x in range(10)])
fig4.show()
In [10]:
fig5 = go.Figure(data=[go.Surface(z=Accuracy_v05)])
fig5.update_layout(title='Accuracy Surface', autosize=True,
scene = dict(xaxis_title='Neurons',
yaxis_title='Layers',
zaxis_title='Accuracy'))
fig5.update_yaxes(tickvals=[x+30 for x in range(10)])
fig5.update_xaxes(tickvals=[x+30 for x in range(10)])
fig5.show()
In [11]:
fig6 = go.Figure(data=[go.Surface(z=Accuracy_v09)])
fig6.update_layout(title='Accuracy Surface', autosize=True,
scene = dict(xaxis_title='Neurons',
yaxis_title='Layers',
zaxis_title='Accuracy'))
fig6.update_yaxes(tickvals=[x+30 for x in range(10)])
fig6.update_xaxes(tickvals=[x+30 for x in range(10)])
fig6.show()
In [12]:
for i in range(10):
for j in range(10):
plt.figure()
plt.title(str(all_combinations[i][j]))
plt.xlabel("Training examples")
plt.ylabel("Score")
plt.grid()
plt.plot(TSZ_LC_V02, TrSCr_LC_V02[i,:,j], 'o-', color="r",
label="Training score")
plt.plot(TSZ_LC_V02, TeSCr_LC_V02[i,:,j], 'o-', color="g",
label="Cross-validation score")
plt.legend(loc="best")
plt.show()
